Matt Arsenault [Thu, 8 Dec 2016 20:14:46 +0000 (20:14 +0000)]
AMDGPU: Make f16 ConstantFP legal
Not having this legal led to combine failures, resulting
in dumb things like bitcasts of constants not being folded
away.
The only reason I'm leaving the v_mov_b32 hack that f32
already uses is to avoid madak formation test regressions.
PeepholeOptimizer has an ordering issue where the immediate
fold attempt is into the sgpr->vgpr copy instead of the actual
use. Running it twice avoids that problem.
Matt Arsenault [Thu, 8 Dec 2016 19:52:38 +0000 (19:52 +0000)]
AMDGPU: Fix commuting v_sub_u16
The correct commutable opcode was set to itself, so this
was simply swapping the operands to commute instead of also
changing the opcode to v_subrev_u16.
Multiple metadata values for records such as opencl.ocl.version, llvm.ident
and similar are created after linking several modules. For some of them, notably
opencl.ocl.version, this creates semantic problem because we cannot tell which
version of OpenCL the composite module conforms.
Moreover, such repetitions of identical values often create a huge list of
unneeded metadata, which grows bitcode size both in memory and stored on disk.
It can go up to several Mb when linked against our OpenCL library. Lastly, such
long lists obscure reading of dumped IR.
IR, X86: Understand !absolute_symbol metadata on global variables.
Summary:
Attaching !absolute_symbol to a global variable does two things:
1) Marks it as an absolute symbol reference.
2) Specifies the value range of that symbol's address.
Teach the X86 backend to allow absolute symbols to appear in place of
immediates by extending the relocImm and mov64imm32 matchers. Start using
relocImm in more places where it is legal.
As previously proposed on llvm-dev:
http://lists.llvm.org/pipermail/llvm-dev/2016-October/105800.html
Summary:
LC can currently select scalar load for uniform memory access
basing on readonly memory address space only. This restriction
originated from the fact that in HW prior to VI vector and scalar caches
are not coherent. With MemoryDependenceAnalysis we can check that the
memory location corresponding to the memory operand of the LOAD is not
clobbered along the all paths from the function entry.
Keno Fischer [Thu, 8 Dec 2016 17:22:35 +0000 (17:22 +0000)]
ConstantFolding: Don't crash when encountering vector GEP
ConstantFolding tried to cast one of the scalar indices to a vector
type. Instead, use the vector type only for the first index (which
is the only one allowed to be a vector) and use its scalar type
otherwise.
Nicolai Haehnle [Thu, 8 Dec 2016 14:08:14 +0000 (14:08 +0000)]
[SelectionDAG] Add expansion and promotion of [US]MUL_LOHI
Summary:
Most targets set the action for these nodes to Expand even though there
isn't actually any code for them in ExpandNode. Instead, targets simply
relied on the fact that no code generates these nodes as long as the
nodes aren't legal or custom.
However, generating these nodes can be useful e.g. for divide-by-constant
in wider integer types.
Expand of [US]MUL_LOHI will use MULH[US] when legal or custom, and
a sequence of half-width multiplications otherwise. Promote uses a wider
multiply.
This patch intends to not change the generated code, but indirect effects
are possible since expansions/promotions that were previously done in
DAGCombine may now be done in LegalizeDAG.
See D24822 for a change that actually uses the new expansion.
Nicolai Haehnle [Thu, 8 Dec 2016 14:08:02 +0000 (14:08 +0000)]
AMDGPU: Properly implement SIRegisterInfo::isFrameOffsetLegal and needsFrameBaseReg
Summary:
Without the fix to isFrameOffsetLegal to consider the instruction's
immediate offset, the new test case hits the corresponding assertion in
resolveFrameIndex, because the LocalStackSlotAllocation pass re-uses a
different base register.
With only the fix to isFrameOffsetLegal, code quality reduces in a bunch of
places because frame base registers are added where they're not needed.
This is addressed by properly implementing needsFrameBaseReg, which also
helps to avoid unnecessary zero frame indices in a bunch of other places.
Daniel Jasper [Thu, 8 Dec 2016 12:45:29 +0000 (12:45 +0000)]
Move DwarfGenerator.cpp to unittests
So far it creates a test helper and so it should be moved there. It also
create a layering cycle between CodeGen and CodeGen/AsmPrinter, which
should be avoided.
Alexey Bataev [Thu, 8 Dec 2016 11:57:51 +0000 (11:57 +0000)]
[SLP] Fix for PR6246: vectorization for scalar ops on vector elements.
When trying to vectorize trees that start at insertelement instructions
function tryToVectorizeList() uses vectorization factor calculated as
MinVecRegSize/ScalarTypeSize. But sometimes it does not work as tree
cost for this fixed vectorization factor is too high.
Patch tries to improve the situation. It tries different vectorization
factors from max(PowerOf2Floor(NumberOfVectorizedValues),
MinVecRegSize/ScalarTypeSize) to MinVecRegSize/ScalarTypeSize and tries
to choose the best one.
Pavel Labath [Thu, 8 Dec 2016 11:31:19 +0000 (11:31 +0000)]
Improve format member detection in llvm::formatv
Summary:
The existing detection of a format member function has a couple of deficiencies:
- the member function does not get detected if one calls formatv with an lvalue,
because the template parameter gets deduced as T&, which fails the is_class
check.
- it also did not work if the function was called with a const variable because
the template parameter would get deduced as const T&, again failing the
is_class check.
This fixes the problem by stripping the references in the uses_format_member
template, to make sure the type is correctly detected as class. It also provides
specializations of the has_FormatMember template for const and non-const members
of the types in order to enable declaring the format member as a "const"
function. I have added tests that verify that formatv can be now called in these
scenarios. As some scenarios could not be verified at runtime (e.g. making sure
that calling a non-const format member on a const object does *not* compile), I
have also added some static_asserts which test the behaviour of the template
classes used internally by formatv().
Keno Fischer [Thu, 8 Dec 2016 01:40:21 +0000 (01:40 +0000)]
[CodeGen] Fix invalid DWARF info on Win64
The relocations for `DIEEntry::EmitValue` were wrong for Win64
(emitting FK_Data_4 instead of FK_SecRel_4). This corrects that
oversight so that the DWARF data is correct in Win64 COFF files.
Fixes PR15393.
Patch by Jameson Nash <jameson@juliacomputing.com> based on a patch
by David Majnemer.
Greg Clayton [Thu, 8 Dec 2016 01:03:48 +0000 (01:03 +0000)]
Make a DWARF generator so we can unit test DWARF APIs with gtest.
The only tests we have for the DWARF parser are the tests that use llvm-dwarfdump and expect output from textual dumps.
More DWARF parser modification are coming in the next few weeks and I wanted to add tests that can verify that we can encode and decode all form types, as well as test some other basic DWARF APIs where we ask DIE objects for their children and siblings.
DwarfGenerator.cpp was added in the lib/CodeGen directory. This file contains the code necessary to easily create DWARF for tests:
StringRef FileBytes = DG.generate();
MemoryBufferRef FileBuffer(FileBytes, "dwarf");
auto Obj = object::ObjectFile::createObjectFile(FileBuffer);
EXPECT_TRUE((bool)Obj);
DWARFContextInMemory DwarfContext(*Obj.get());
This code is backed by the AsmPrinter code that emits DWARF for the actual compiler.
While adding unit tests it was discovered that DIEValue that used DIEEntry as their values had bugs where DW_FORM_ref1, DW_FORM_ref2, DW_FORM_ref8, and DW_FORM_ref_udata forms were not supported. These are all now supported. Added support for DW_FORM_string so we can emit inlined C strings.
Centralized the code to unique abbreviations into a new DIEAbbrevSet class and made both the dwarfgen::Generator and the llvm::DwarfFile classes use the new class.
Fixed comments in the llvm::DIE class so that the Offset is known to be the compile/type unit offset.
DIEInteger now supports more DW_FORM values.
There are also unit tests that cover:
Encoding and decoding all form types and values
Encoding and decoding all reference types (DW_FORM_ref1, DW_FORM_ref2, DW_FORM_ref4, DW_FORM_ref8, DW_FORM_ref_udata, DW_FORM_ref_addr) including cross compile unit references with that go forward one compile unit and backward on compile unit.
Replace @progbits in the section directive with %progbits, because "@" starts a comment on arm/thumb.
Use b.w branch instruction.
Use .thumb_function and .thumb_set for proper arm/thumb interwork. This way jumptable entry addresses on thumb have bit 0 set (correctly). This does not affect CFI check math, because the address of the jumptable start also has that bit set.
This does not work on thumbv5, because it does not support b.w, and the linker would not insert a veneer (trampoline?) to extend the range of b.n. We may need to do full-range plt-style jumptables on thumbv54, which are 12 bytes per entry. Another option is "push lr; bl; pop pc" (4 bytes) but that needs unwinding instructions, etc.
Davide Italiano [Wed, 7 Dec 2016 21:47:32 +0000 (21:47 +0000)]
[BDCE] Skip metadata while replacing uses.
The fix committed in r288851 doesn't cover all the cases.
In particular, if we have an instruction with side effects
which has a no non-dbg use not depending on the bits, we still
perform RAUW destroying the dbg.value's first argument.
Prevent metadata from being replaced here to avoid the issue.
Tim Northover [Wed, 7 Dec 2016 21:29:15 +0000 (21:29 +0000)]
GlobalISel: use correct builder for ConstantExprs.
ConstantExpr instances were emitting code into the current block rather than
the entry block. This meant they didn't necessarily dominate all uses, which is
clearly wrong.
Chris Bieneman [Wed, 7 Dec 2016 21:26:32 +0000 (21:26 +0000)]
[ObjectYAML] Pull DWARF support into DWARFYAML namespace
Since DWARF formatting is agnostic to the object file it is stored in, it doesn't make sense for this to be in the MachOYAML implementation. Pulling it into its own namespace means we could modify the ELF and COFF YAML tools to emit DWARF as well.
In a follow-up patch I will better abstract this in obj2yaml and yaml2obj so that the DWARF bits in the tools can be re-used too.
Tim Northover [Wed, 7 Dec 2016 21:05:38 +0000 (21:05 +0000)]
GlobalISel: simplify MachineIRBuilder interface.
MachineIRBuilder had weird before/after and beginning/end flags for the insert
point. Unfortunately the non-default means that instructions will be inserted
in reverse order which is almost never what anyone wants.
Really, I think we just want (like IRBuilder has) the ability to insert at any
C++ iterator-style point (i.e. before any instruction or before MBB.end()). So
this fixes MIRBuilders to behave like IRBuilders in this respect.
Sanjay Patel [Wed, 7 Dec 2016 18:56:55 +0000 (18:56 +0000)]
[InstCombine] add tests for smin+icmp; NFC
The tests that already work are folded in InstSimplify, so those
tests should be redundant and we can remove them if they don't
seem worthwhile for completeness.
Matthew Simpson [Wed, 7 Dec 2016 15:03:32 +0000 (15:03 +0000)]
[LV] Scalarize operands of predicated instructions
This patch attempts to scalarize the operand expressions of predicated
instructions if they were conditionally executed in the original loop. After
scalarization, the expressions will be sunk inside the blocks created for the
predicated instructions. The transformation essentially performs
un-if-conversion on the operands.
The cost model has been updated to determine if scalarization is profitable. It
compares the cost of a vectorized instruction, assuming it will be
if-converted, to the cost of the scalarized instruction, assuming that the
instructions corresponding to each vector lane will be sunk inside a predicated
block, possibly avoiding execution. If it's more profitable to scalarize the
entire expression tree feeding the predicated instruction, the expression will
be scalarized; otherwise, it will be vectorized. We only consider the cost of
the entire expression to accurately estimate the cost of the required
insertelement and extractelement instructions.
When GVN removes a redundant load, it should not modify the debug location of the dominating load.
In the case of a fully redundant load LI dominated by an equivalent load V, GVN
should always preserve the original debug location of V. Otherwise, we risk to
introduce an incorrect stepping.
If V has debug info, then clearly it should not be modified. If V has a null
debugloc, then it is still potentially incorrect to propagate LI's debugloc
because LI may not post-dominate V.
Simon Pilgrim [Wed, 7 Dec 2016 11:19:00 +0000 (11:19 +0000)]
[X86][XOP] Fix VPERMIL2 non-constant pool shuffle decoding (PR31296)
The non-constant pool version of DecodeVPERMIL2PMask was not offsetting correctly for the second input. I've updated the code to match the implementation in the constant-pool version.
Annoyingly this bug was hidden for so long as it's tricky to combine to useful variable shuffle masks that don't become constant-pool entries.
[InlineFunction] Do not propagate the callsite debug location to instructions inlined from functions with debug info.
When a function F is inlined, InlineFunction extends the debug location of every
instruction inlined from F by adding an InlinedAt.
However, if an instruction has a 'null' debug location, InlineFunction would
propagate the callsite debug location to it. This behavior existed since
revision 210459.
Revision 210459 was originally committed specifically to workaround the lack of
debug information for instructions inlined from intrinsic functions (which are
usually declared with attributes `__always_inline__, __nodebug__`).
The problem with revision 210459 is that it doesn't make any sort of distinction
between instructions inlined from a 'nodebug' function and instructions which
are inlined from a function built with debug info. This issue may lead to
incorrect stepping in the debugger.
This patch works under the assumption that a nodebug function does not have a
DISubprogram. When a function F is inlined into another function G,
InlineFunction checks if F has debug info associated with it.
For nodebug functions, the InlineFunction logic is unchanged (i.e. it would
still propagate the callsite debugloc to the inlined instructions). Otherwise,
InlineFunction no longer propagates the callsite debug location.
Philip Reames [Wed, 7 Dec 2016 04:48:50 +0000 (04:48 +0000)]
Reintroduce a check accidentally removed in 288873 to fix clang bots
I believe this is the cause of the failure, but have not been able to confirm. Note that this is a speculative fix; I'm still waiting for a full build to finish as I synced and ended up doing a clean build which takes 20+ minutes on my machine.
LowerTypeTests: Improve performance by optimising type metadata queries.
Requesting metadata for a global is a relatively expensive operation as it
involves a map lookup, but it's one that we need to do relatively frequently in
this pass to collect the list of type metadata nodes associated with a global.
This change improves the performance of type metadata queries by prebuilding
data structures that keep the global together with its list of type metadata,
and changing the pass to use that data structure wherever we were previously
passing global references around.
This change also eliminates some O(N^2) behavior by collecting the list of
globals associated with each type identifier during the first pass over the
list of globals rather than visiting each global to compute that list every
time we add a new type identifier.
Reduces pass runtime on a module containing Chrome's vtables from over 60s
to 0.9s.
projects / llvm / log